TACC’s Stampede3 Supercomputer Makes use of Intel’s Xeon Max with HBM2E and Ponte Vecchio
The Texas Superior Computing Middle (TACC) unveiled its newest Stampede supercomputer for open science analysis tasks, Stampede3. TACC anticipates that Stampede3 will come on-line this fall and can ship its full efficiency in early 2024. The supercomputer might be an important element of the U.S. Nationwide Science Basis’s (NSF) ACCESS scientific supercomputing ecosystem, and it’s projected to serve the open science group from 2024 till 2029.
The third-generation Stampede cluster, which might be constructed by Dell, will incorporate 560 nodes geared up with Intel’s Sapphire Rapids era Xeon CPU Max processors, every providing 56 CPU cores and 64GB of on-package HBM2E reminiscence. Surprisingly, TACC goes to be working these nodes in HBM-only mode, so no extra DRAM might be hooked up to the CPU nodes – all of their reminiscence will come from the on-chip HBM stacks.
With these specs, Stampede3 is anticipated to have a peak efficiency of roughly 4 FP64 PetaFLOPS, whereas providing practically 63,000 general-purpose cores. As well as, TACC additionally plans to put in 10 Dell PowerEdge XE9640 servers with 40 Intel Information Middle GPU Max compute GPUs for synthetic intelligence and machine studying workloads.
Given this format, the majority of Stampede3’s compute efficiency might be equipped by CPUs. This makes Stampede3 a little bit of a rarity at the present time, as most high-performance techniques are GPU pushed, leaving Stampede3 as one of many final supercomputers that depends virtually solely on general-purpose CPUs.
And whereas the present cluster is primarily targeted on CPU efficiency, TACC can be going to make use of the Intel GPUs within the newest Stampede revamp to research on incorporate bigger numbers of GPUs into future variations of the system. For now, most of TACC’s AI duties are run on its Lone Star techniques, which is powered by lots of Nvidia A100 compute GPUs. So the group’s intention is to discover whether or not a portion of this workload might be transferred to Intel’s Ponte Vecchio.
We’re going to put in a small system with exploratory functionality utilizing Intel Ponte Vecchio,” mentioned Dan Stanzione, government director of TACC. “We’re nonetheless negotiating precisely how a lot of that can have, however I might say a minimal of 40 nodes and most of 100 or so. […] We’re simply placing a few racks of Ponte Vecchio on the market to see how individuals work with it.”
Stampede3 will leverage 400 Gb/s Omni-Path Material expertise that can allow a backplane bandwidth of 24TB/s. This setup will permit the machine to effectively scale and decrease latencies, making it well-suited for numerous functions requiring simulations.
TACC additionally plans to reincorporate nodes from the earlier model, Stampede2, which have been based mostly on older-generation Xeon Scalable CPUs. This integration will improve the capability of Stampede3 for high-memory functions, high-throughput computing, interactive workloads, and different previous-generation functions. In whole, the brand new supercomputer system will function 1,858 compute nodes with over 140,000 cores, greater than 330 TBs of RAM, new storage capability of 13 PBs, and a peak efficiency near 10 PetaFLOPS.
